Ignorability for general longitudinal data

Area of Science:

• Causal inference
• Longitudinal data analysis
• Statistical modeling

Background:

• Identifying causal effects is crucial in many scientific fields.
• Covariate adjustment is a common method for causal inference.
• The concept of ignorability is key to valid causal effect estimation.

Purpose of the Study:

• To demonstrate the link between ignorability and causal effect identification via covariate adjustment.
• To introduce a graphical condition, stability, for assessing ignorability in longitudinal data.
• To provide a formulation of ignorability not reliant on missing data concepts.

Main Methods:

• Developing a graphical condition termed stability.
• Applying stability to general longitudinal data.
• Illustrating stability assessment with examples.

Main Results:

• Established a close relationship between ignorability and causal effect identification through covariate adjustment.
• Introduced stability as a graphical condition analogous to missingness at random for longitudinal data.
• Showcased the applicability of stability in scenarios without actual missing data.

Conclusions:

• Ignorability and causal effect identification are tightly linked, particularly with covariate adjustment.
• Stability offers a robust graphical criterion for ignorability in longitudinal studies, irrespective of missing data.
• The stability condition provides a practical tool for assessing assumptions in causal inference for longitudinal data.